1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 - Virtual Open Systems and Columbia University 4 * Author: Christoffer Dall <c.dall@virtualopensystems.com> 5 */ 6 7 #include <linux/bug.h> 8 #include <linux/cpu_pm.h> 9 #include <linux/entry-kvm.h> 10 #include <linux/errno.h> 11 #include <linux/err.h> 12 #include <linux/kvm_host.h> 13 #include <linux/list.h> 14 #include <linux/module.h> 15 #include <linux/vmalloc.h> 16 #include <linux/fs.h> 17 #include <linux/mman.h> 18 #include <linux/sched.h> 19 #include <linux/kvm.h> 20 #include <linux/kvm_irqfd.h> 21 #include <linux/irqbypass.h> 22 #include <linux/sched/stat.h> 23 #include <linux/psci.h> 24 #include <trace/events/kvm.h> 25 26 #define CREATE_TRACE_POINTS 27 #include "trace_arm.h" 28 29 #include <linux/uaccess.h> 30 #include <asm/ptrace.h> 31 #include <asm/mman.h> 32 #include <asm/tlbflush.h> 33 #include <asm/cacheflush.h> 34 #include <asm/cpufeature.h> 35 #include <asm/virt.h> 36 #include <asm/kvm_arm.h> 37 #include <asm/kvm_asm.h> 38 #include <asm/kvm_mmu.h> 39 #include <asm/kvm_nested.h> 40 #include <asm/kvm_pkvm.h> 41 #include <asm/kvm_emulate.h> 42 #include <asm/sections.h> 43 44 #include <kvm/arm_hypercalls.h> 45 #include <kvm/arm_pmu.h> 46 #include <kvm/arm_psci.h> 47 48 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT; 49 50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector); 51 52 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); 53 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params); 54 55 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt); 56 57 static bool vgic_present, kvm_arm_initialised; 58 59 static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized); 60 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); 61 62 bool is_kvm_arm_initialised(void) 63 { 64 return kvm_arm_initialised; 65 } 66 67 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 68 { 69 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 70 } 71 72 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 73 struct kvm_enable_cap *cap) 74 { 75 int r; 76 u64 new_cap; 77 78 if (cap->flags) 79 return -EINVAL; 80 81 switch (cap->cap) { 82 case KVM_CAP_ARM_NISV_TO_USER: 83 r = 0; 84 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER, 85 &kvm->arch.flags); 86 break; 87 case KVM_CAP_ARM_MTE: 88 mutex_lock(&kvm->lock); 89 if (!system_supports_mte() || kvm->created_vcpus) { 90 r = -EINVAL; 91 } else { 92 r = 0; 93 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags); 94 } 95 mutex_unlock(&kvm->lock); 96 break; 97 case KVM_CAP_ARM_SYSTEM_SUSPEND: 98 r = 0; 99 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags); 100 break; 101 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE: 102 new_cap = cap->args[0]; 103 104 mutex_lock(&kvm->slots_lock); 105 /* 106 * To keep things simple, allow changing the chunk 107 * size only when no memory slots have been created. 108 */ 109 if (!kvm_are_all_memslots_empty(kvm)) { 110 r = -EINVAL; 111 } else if (new_cap && !kvm_is_block_size_supported(new_cap)) { 112 r = -EINVAL; 113 } else { 114 r = 0; 115 kvm->arch.mmu.split_page_chunk_size = new_cap; 116 } 117 mutex_unlock(&kvm->slots_lock); 118 break; 119 default: 120 r = -EINVAL; 121 break; 122 } 123 124 return r; 125 } 126 127 static int kvm_arm_default_max_vcpus(void) 128 { 129 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; 130 } 131 132 /** 133 * kvm_arch_init_vm - initializes a VM data structure 134 * @kvm: pointer to the KVM struct 135 */ 136 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 137 { 138 int ret; 139 140 mutex_init(&kvm->arch.config_lock); 141 142 #ifdef CONFIG_LOCKDEP 143 /* Clue in lockdep that the config_lock must be taken inside kvm->lock */ 144 mutex_lock(&kvm->lock); 145 mutex_lock(&kvm->arch.config_lock); 146 mutex_unlock(&kvm->arch.config_lock); 147 mutex_unlock(&kvm->lock); 148 #endif 149 150 ret = kvm_share_hyp(kvm, kvm + 1); 151 if (ret) 152 return ret; 153 154 ret = pkvm_init_host_vm(kvm); 155 if (ret) 156 goto err_unshare_kvm; 157 158 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) { 159 ret = -ENOMEM; 160 goto err_unshare_kvm; 161 } 162 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask); 163 164 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type); 165 if (ret) 166 goto err_free_cpumask; 167 168 kvm_vgic_early_init(kvm); 169 170 kvm_timer_init_vm(kvm); 171 172 /* The maximum number of VCPUs is limited by the host's GIC model */ 173 kvm->max_vcpus = kvm_arm_default_max_vcpus(); 174 175 kvm_arm_init_hypercalls(kvm); 176 177 bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES); 178 179 return 0; 180 181 err_free_cpumask: 182 free_cpumask_var(kvm->arch.supported_cpus); 183 err_unshare_kvm: 184 kvm_unshare_hyp(kvm, kvm + 1); 185 return ret; 186 } 187 188 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 189 { 190 return VM_FAULT_SIGBUS; 191 } 192 193 194 /** 195 * kvm_arch_destroy_vm - destroy the VM data structure 196 * @kvm: pointer to the KVM struct 197 */ 198 void kvm_arch_destroy_vm(struct kvm *kvm) 199 { 200 bitmap_free(kvm->arch.pmu_filter); 201 free_cpumask_var(kvm->arch.supported_cpus); 202 203 kvm_vgic_destroy(kvm); 204 205 if (is_protected_kvm_enabled()) 206 pkvm_destroy_hyp_vm(kvm); 207 208 kvm_destroy_vcpus(kvm); 209 210 kvm_unshare_hyp(kvm, kvm + 1); 211 212 kvm_arm_teardown_hypercalls(kvm); 213 } 214 215 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 216 { 217 int r; 218 switch (ext) { 219 case KVM_CAP_IRQCHIP: 220 r = vgic_present; 221 break; 222 case KVM_CAP_IOEVENTFD: 223 case KVM_CAP_DEVICE_CTRL: 224 case KVM_CAP_USER_MEMORY: 225 case KVM_CAP_SYNC_MMU: 226 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 227 case KVM_CAP_ONE_REG: 228 case KVM_CAP_ARM_PSCI: 229 case KVM_CAP_ARM_PSCI_0_2: 230 case KVM_CAP_READONLY_MEM: 231 case KVM_CAP_MP_STATE: 232 case KVM_CAP_IMMEDIATE_EXIT: 233 case KVM_CAP_VCPU_EVENTS: 234 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: 235 case KVM_CAP_ARM_NISV_TO_USER: 236 case KVM_CAP_ARM_INJECT_EXT_DABT: 237 case KVM_CAP_SET_GUEST_DEBUG: 238 case KVM_CAP_VCPU_ATTRIBUTES: 239 case KVM_CAP_PTP_KVM: 240 case KVM_CAP_ARM_SYSTEM_SUSPEND: 241 case KVM_CAP_IRQFD_RESAMPLE: 242 case KVM_CAP_COUNTER_OFFSET: 243 r = 1; 244 break; 245 case KVM_CAP_SET_GUEST_DEBUG2: 246 return KVM_GUESTDBG_VALID_MASK; 247 case KVM_CAP_ARM_SET_DEVICE_ADDR: 248 r = 1; 249 break; 250 case KVM_CAP_NR_VCPUS: 251 /* 252 * ARM64 treats KVM_CAP_NR_CPUS differently from all other 253 * architectures, as it does not always bound it to 254 * KVM_CAP_MAX_VCPUS. It should not matter much because 255 * this is just an advisory value. 256 */ 257 r = min_t(unsigned int, num_online_cpus(), 258 kvm_arm_default_max_vcpus()); 259 break; 260 case KVM_CAP_MAX_VCPUS: 261 case KVM_CAP_MAX_VCPU_ID: 262 if (kvm) 263 r = kvm->max_vcpus; 264 else 265 r = kvm_arm_default_max_vcpus(); 266 break; 267 case KVM_CAP_MSI_DEVID: 268 if (!kvm) 269 r = -EINVAL; 270 else 271 r = kvm->arch.vgic.msis_require_devid; 272 break; 273 case KVM_CAP_ARM_USER_IRQ: 274 /* 275 * 1: EL1_VTIMER, EL1_PTIMER, and PMU. 276 * (bump this number if adding more devices) 277 */ 278 r = 1; 279 break; 280 case KVM_CAP_ARM_MTE: 281 r = system_supports_mte(); 282 break; 283 case KVM_CAP_STEAL_TIME: 284 r = kvm_arm_pvtime_supported(); 285 break; 286 case KVM_CAP_ARM_EL1_32BIT: 287 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1); 288 break; 289 case KVM_CAP_GUEST_DEBUG_HW_BPS: 290 r = get_num_brps(); 291 break; 292 case KVM_CAP_GUEST_DEBUG_HW_WPS: 293 r = get_num_wrps(); 294 break; 295 case KVM_CAP_ARM_PMU_V3: 296 r = kvm_arm_support_pmu_v3(); 297 break; 298 case KVM_CAP_ARM_INJECT_SERROR_ESR: 299 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN); 300 break; 301 case KVM_CAP_ARM_VM_IPA_SIZE: 302 r = get_kvm_ipa_limit(); 303 break; 304 case KVM_CAP_ARM_SVE: 305 r = system_supports_sve(); 306 break; 307 case KVM_CAP_ARM_PTRAUTH_ADDRESS: 308 case KVM_CAP_ARM_PTRAUTH_GENERIC: 309 r = system_has_full_ptr_auth(); 310 break; 311 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE: 312 if (kvm) 313 r = kvm->arch.mmu.split_page_chunk_size; 314 else 315 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT; 316 break; 317 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES: 318 r = kvm_supported_block_sizes(); 319 break; 320 default: 321 r = 0; 322 } 323 324 return r; 325 } 326 327 long kvm_arch_dev_ioctl(struct file *filp, 328 unsigned int ioctl, unsigned long arg) 329 { 330 return -EINVAL; 331 } 332 333 struct kvm *kvm_arch_alloc_vm(void) 334 { 335 size_t sz = sizeof(struct kvm); 336 337 if (!has_vhe()) 338 return kzalloc(sz, GFP_KERNEL_ACCOUNT); 339 340 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO); 341 } 342 343 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 344 { 345 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) 346 return -EBUSY; 347 348 if (id >= kvm->max_vcpus) 349 return -EINVAL; 350 351 return 0; 352 } 353 354 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 355 { 356 int err; 357 358 spin_lock_init(&vcpu->arch.mp_state_lock); 359 360 #ifdef CONFIG_LOCKDEP 361 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */ 362 mutex_lock(&vcpu->mutex); 363 mutex_lock(&vcpu->kvm->arch.config_lock); 364 mutex_unlock(&vcpu->kvm->arch.config_lock); 365 mutex_unlock(&vcpu->mutex); 366 #endif 367 368 /* Force users to call KVM_ARM_VCPU_INIT */ 369 vcpu_clear_flag(vcpu, VCPU_INITIALIZED); 370 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 371 372 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 373 374 /* 375 * Default value for the FP state, will be overloaded at load 376 * time if we support FP (pretty likely) 377 */ 378 vcpu->arch.fp_state = FP_STATE_FREE; 379 380 /* Set up the timer */ 381 kvm_timer_vcpu_init(vcpu); 382 383 kvm_pmu_vcpu_init(vcpu); 384 385 kvm_arm_reset_debug_ptr(vcpu); 386 387 kvm_arm_pvtime_vcpu_init(&vcpu->arch); 388 389 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu; 390 391 err = kvm_vgic_vcpu_init(vcpu); 392 if (err) 393 return err; 394 395 return kvm_share_hyp(vcpu, vcpu + 1); 396 } 397 398 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 399 { 400 } 401 402 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 403 { 404 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm))) 405 static_branch_dec(&userspace_irqchip_in_use); 406 407 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 408 kvm_timer_vcpu_terminate(vcpu); 409 kvm_pmu_vcpu_destroy(vcpu); 410 kvm_vgic_vcpu_destroy(vcpu); 411 kvm_arm_vcpu_destroy(vcpu); 412 } 413 414 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) 415 { 416 417 } 418 419 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) 420 { 421 422 } 423 424 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 425 { 426 struct kvm_s2_mmu *mmu; 427 int *last_ran; 428 429 mmu = vcpu->arch.hw_mmu; 430 last_ran = this_cpu_ptr(mmu->last_vcpu_ran); 431 432 /* 433 * We guarantee that both TLBs and I-cache are private to each 434 * vcpu. If detecting that a vcpu from the same VM has 435 * previously run on the same physical CPU, call into the 436 * hypervisor code to nuke the relevant contexts. 437 * 438 * We might get preempted before the vCPU actually runs, but 439 * over-invalidation doesn't affect correctness. 440 */ 441 if (*last_ran != vcpu->vcpu_id) { 442 kvm_call_hyp(__kvm_flush_cpu_context, mmu); 443 *last_ran = vcpu->vcpu_id; 444 } 445 446 vcpu->cpu = cpu; 447 448 kvm_vgic_load(vcpu); 449 kvm_timer_vcpu_load(vcpu); 450 if (has_vhe()) 451 kvm_vcpu_load_sysregs_vhe(vcpu); 452 kvm_arch_vcpu_load_fp(vcpu); 453 kvm_vcpu_pmu_restore_guest(vcpu); 454 if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) 455 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); 456 457 if (single_task_running()) 458 vcpu_clear_wfx_traps(vcpu); 459 else 460 vcpu_set_wfx_traps(vcpu); 461 462 if (vcpu_has_ptrauth(vcpu)) 463 vcpu_ptrauth_disable(vcpu); 464 kvm_arch_vcpu_load_debug_state_flags(vcpu); 465 466 if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus)) 467 vcpu_set_on_unsupported_cpu(vcpu); 468 } 469 470 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 471 { 472 kvm_arch_vcpu_put_debug_state_flags(vcpu); 473 kvm_arch_vcpu_put_fp(vcpu); 474 if (has_vhe()) 475 kvm_vcpu_put_sysregs_vhe(vcpu); 476 kvm_timer_vcpu_put(vcpu); 477 kvm_vgic_put(vcpu); 478 kvm_vcpu_pmu_restore_host(vcpu); 479 kvm_arm_vmid_clear_active(); 480 481 vcpu_clear_on_unsupported_cpu(vcpu); 482 vcpu->cpu = -1; 483 } 484 485 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu) 486 { 487 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED); 488 kvm_make_request(KVM_REQ_SLEEP, vcpu); 489 kvm_vcpu_kick(vcpu); 490 } 491 492 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu) 493 { 494 spin_lock(&vcpu->arch.mp_state_lock); 495 __kvm_arm_vcpu_power_off(vcpu); 496 spin_unlock(&vcpu->arch.mp_state_lock); 497 } 498 499 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu) 500 { 501 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED; 502 } 503 504 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu) 505 { 506 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED); 507 kvm_make_request(KVM_REQ_SUSPEND, vcpu); 508 kvm_vcpu_kick(vcpu); 509 } 510 511 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu) 512 { 513 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED; 514 } 515 516 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 517 struct kvm_mp_state *mp_state) 518 { 519 *mp_state = READ_ONCE(vcpu->arch.mp_state); 520 521 return 0; 522 } 523 524 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 525 struct kvm_mp_state *mp_state) 526 { 527 int ret = 0; 528 529 spin_lock(&vcpu->arch.mp_state_lock); 530 531 switch (mp_state->mp_state) { 532 case KVM_MP_STATE_RUNNABLE: 533 WRITE_ONCE(vcpu->arch.mp_state, *mp_state); 534 break; 535 case KVM_MP_STATE_STOPPED: 536 __kvm_arm_vcpu_power_off(vcpu); 537 break; 538 case KVM_MP_STATE_SUSPENDED: 539 kvm_arm_vcpu_suspend(vcpu); 540 break; 541 default: 542 ret = -EINVAL; 543 } 544 545 spin_unlock(&vcpu->arch.mp_state_lock); 546 547 return ret; 548 } 549 550 /** 551 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled 552 * @v: The VCPU pointer 553 * 554 * If the guest CPU is not waiting for interrupts or an interrupt line is 555 * asserted, the CPU is by definition runnable. 556 */ 557 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 558 { 559 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); 560 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) 561 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause); 562 } 563 564 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 565 { 566 return vcpu_mode_priv(vcpu); 567 } 568 569 #ifdef CONFIG_GUEST_PERF_EVENTS 570 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu) 571 { 572 return *vcpu_pc(vcpu); 573 } 574 #endif 575 576 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) 577 { 578 return vcpu_get_flag(vcpu, VCPU_INITIALIZED); 579 } 580 581 /* 582 * Handle both the initialisation that is being done when the vcpu is 583 * run for the first time, as well as the updates that must be 584 * performed each time we get a new thread dealing with this vcpu. 585 */ 586 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu) 587 { 588 struct kvm *kvm = vcpu->kvm; 589 int ret; 590 591 if (!kvm_vcpu_initialized(vcpu)) 592 return -ENOEXEC; 593 594 if (!kvm_arm_vcpu_is_finalized(vcpu)) 595 return -EPERM; 596 597 ret = kvm_arch_vcpu_run_map_fp(vcpu); 598 if (ret) 599 return ret; 600 601 if (likely(vcpu_has_run_once(vcpu))) 602 return 0; 603 604 kvm_arm_vcpu_init_debug(vcpu); 605 606 if (likely(irqchip_in_kernel(kvm))) { 607 /* 608 * Map the VGIC hardware resources before running a vcpu the 609 * first time on this VM. 610 */ 611 ret = kvm_vgic_map_resources(kvm); 612 if (ret) 613 return ret; 614 } 615 616 ret = kvm_timer_enable(vcpu); 617 if (ret) 618 return ret; 619 620 ret = kvm_arm_pmu_v3_enable(vcpu); 621 if (ret) 622 return ret; 623 624 if (is_protected_kvm_enabled()) { 625 ret = pkvm_create_hyp_vm(kvm); 626 if (ret) 627 return ret; 628 } 629 630 if (!irqchip_in_kernel(kvm)) { 631 /* 632 * Tell the rest of the code that there are userspace irqchip 633 * VMs in the wild. 634 */ 635 static_branch_inc(&userspace_irqchip_in_use); 636 } 637 638 /* 639 * Initialize traps for protected VMs. 640 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once 641 * the code is in place for first run initialization at EL2. 642 */ 643 if (kvm_vm_is_protected(kvm)) 644 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu); 645 646 mutex_lock(&kvm->arch.config_lock); 647 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags); 648 mutex_unlock(&kvm->arch.config_lock); 649 650 return ret; 651 } 652 653 bool kvm_arch_intc_initialized(struct kvm *kvm) 654 { 655 return vgic_initialized(kvm); 656 } 657 658 void kvm_arm_halt_guest(struct kvm *kvm) 659 { 660 unsigned long i; 661 struct kvm_vcpu *vcpu; 662 663 kvm_for_each_vcpu(i, vcpu, kvm) 664 vcpu->arch.pause = true; 665 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); 666 } 667 668 void kvm_arm_resume_guest(struct kvm *kvm) 669 { 670 unsigned long i; 671 struct kvm_vcpu *vcpu; 672 673 kvm_for_each_vcpu(i, vcpu, kvm) { 674 vcpu->arch.pause = false; 675 __kvm_vcpu_wake_up(vcpu); 676 } 677 } 678 679 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu) 680 { 681 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 682 683 rcuwait_wait_event(wait, 684 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause), 685 TASK_INTERRUPTIBLE); 686 687 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) { 688 /* Awaken to handle a signal, request we sleep again later. */ 689 kvm_make_request(KVM_REQ_SLEEP, vcpu); 690 } 691 692 /* 693 * Make sure we will observe a potential reset request if we've 694 * observed a change to the power state. Pairs with the smp_wmb() in 695 * kvm_psci_vcpu_on(). 696 */ 697 smp_rmb(); 698 } 699 700 /** 701 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior 702 * @vcpu: The VCPU pointer 703 * 704 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until 705 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending 706 * on when a wake event arrives, e.g. there may already be a pending wake event. 707 */ 708 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu) 709 { 710 /* 711 * Sync back the state of the GIC CPU interface so that we have 712 * the latest PMR and group enables. This ensures that 713 * kvm_arch_vcpu_runnable has up-to-date data to decide whether 714 * we have pending interrupts, e.g. when determining if the 715 * vCPU should block. 716 * 717 * For the same reason, we want to tell GICv4 that we need 718 * doorbells to be signalled, should an interrupt become pending. 719 */ 720 preempt_disable(); 721 kvm_vgic_vmcr_sync(vcpu); 722 vcpu_set_flag(vcpu, IN_WFI); 723 vgic_v4_put(vcpu); 724 preempt_enable(); 725 726 kvm_vcpu_halt(vcpu); 727 vcpu_clear_flag(vcpu, IN_WFIT); 728 729 preempt_disable(); 730 vcpu_clear_flag(vcpu, IN_WFI); 731 vgic_v4_load(vcpu); 732 preempt_enable(); 733 } 734 735 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu) 736 { 737 if (!kvm_arm_vcpu_suspended(vcpu)) 738 return 1; 739 740 kvm_vcpu_wfi(vcpu); 741 742 /* 743 * The suspend state is sticky; we do not leave it until userspace 744 * explicitly marks the vCPU as runnable. Request that we suspend again 745 * later. 746 */ 747 kvm_make_request(KVM_REQ_SUSPEND, vcpu); 748 749 /* 750 * Check to make sure the vCPU is actually runnable. If so, exit to 751 * userspace informing it of the wakeup condition. 752 */ 753 if (kvm_arch_vcpu_runnable(vcpu)) { 754 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event)); 755 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP; 756 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; 757 return 0; 758 } 759 760 /* 761 * Otherwise, we were unblocked to process a different event, such as a 762 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to 763 * process the event. 764 */ 765 return 1; 766 } 767 768 /** 769 * check_vcpu_requests - check and handle pending vCPU requests 770 * @vcpu: the VCPU pointer 771 * 772 * Return: 1 if we should enter the guest 773 * 0 if we should exit to userspace 774 * < 0 if we should exit to userspace, where the return value indicates 775 * an error 776 */ 777 static int check_vcpu_requests(struct kvm_vcpu *vcpu) 778 { 779 if (kvm_request_pending(vcpu)) { 780 if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) 781 return -EIO; 782 783 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) 784 kvm_vcpu_sleep(vcpu); 785 786 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 787 kvm_reset_vcpu(vcpu); 788 789 /* 790 * Clear IRQ_PENDING requests that were made to guarantee 791 * that a VCPU sees new virtual interrupts. 792 */ 793 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); 794 795 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) 796 kvm_update_stolen_time(vcpu); 797 798 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) { 799 /* The distributor enable bits were changed */ 800 preempt_disable(); 801 vgic_v4_put(vcpu); 802 vgic_v4_load(vcpu); 803 preempt_enable(); 804 } 805 806 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu)) 807 kvm_pmu_handle_pmcr(vcpu, 808 __vcpu_sys_reg(vcpu, PMCR_EL0)); 809 810 if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu)) 811 kvm_vcpu_pmu_restore_guest(vcpu); 812 813 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu)) 814 return kvm_vcpu_suspend(vcpu); 815 816 if (kvm_dirty_ring_check_request(vcpu)) 817 return 0; 818 } 819 820 return 1; 821 } 822 823 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu) 824 { 825 if (likely(!vcpu_mode_is_32bit(vcpu))) 826 return false; 827 828 if (vcpu_has_nv(vcpu)) 829 return true; 830 831 return !kvm_supports_32bit_el0(); 832 } 833 834 /** 835 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest 836 * @vcpu: The VCPU pointer 837 * @ret: Pointer to write optional return code 838 * 839 * Returns: true if the VCPU needs to return to a preemptible + interruptible 840 * and skip guest entry. 841 * 842 * This function disambiguates between two different types of exits: exits to a 843 * preemptible + interruptible kernel context and exits to userspace. For an 844 * exit to userspace, this function will write the return code to ret and return 845 * true. For an exit to preemptible + interruptible kernel context (i.e. check 846 * for pending work and re-enter), return true without writing to ret. 847 */ 848 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret) 849 { 850 struct kvm_run *run = vcpu->run; 851 852 /* 853 * If we're using a userspace irqchip, then check if we need 854 * to tell a userspace irqchip about timer or PMU level 855 * changes and if so, exit to userspace (the actual level 856 * state gets updated in kvm_timer_update_run and 857 * kvm_pmu_update_run below). 858 */ 859 if (static_branch_unlikely(&userspace_irqchip_in_use)) { 860 if (kvm_timer_should_notify_user(vcpu) || 861 kvm_pmu_should_notify_user(vcpu)) { 862 *ret = -EINTR; 863 run->exit_reason = KVM_EXIT_INTR; 864 return true; 865 } 866 } 867 868 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) { 869 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 870 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED; 871 run->fail_entry.cpu = smp_processor_id(); 872 *ret = 0; 873 return true; 874 } 875 876 return kvm_request_pending(vcpu) || 877 xfer_to_guest_mode_work_pending(); 878 } 879 880 /* 881 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while 882 * the vCPU is running. 883 * 884 * This must be noinstr as instrumentation may make use of RCU, and this is not 885 * safe during the EQS. 886 */ 887 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu) 888 { 889 int ret; 890 891 guest_state_enter_irqoff(); 892 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu); 893 guest_state_exit_irqoff(); 894 895 return ret; 896 } 897 898 /** 899 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code 900 * @vcpu: The VCPU pointer 901 * 902 * This function is called through the VCPU_RUN ioctl called from user space. It 903 * will execute VM code in a loop until the time slice for the process is used 904 * or some emulation is needed from user space in which case the function will 905 * return with return value 0 and with the kvm_run structure filled in with the 906 * required data for the requested emulation. 907 */ 908 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 909 { 910 struct kvm_run *run = vcpu->run; 911 int ret; 912 913 if (run->exit_reason == KVM_EXIT_MMIO) { 914 ret = kvm_handle_mmio_return(vcpu); 915 if (ret) 916 return ret; 917 } 918 919 vcpu_load(vcpu); 920 921 if (run->immediate_exit) { 922 ret = -EINTR; 923 goto out; 924 } 925 926 kvm_sigset_activate(vcpu); 927 928 ret = 1; 929 run->exit_reason = KVM_EXIT_UNKNOWN; 930 run->flags = 0; 931 while (ret > 0) { 932 /* 933 * Check conditions before entering the guest 934 */ 935 ret = xfer_to_guest_mode_handle_work(vcpu); 936 if (!ret) 937 ret = 1; 938 939 if (ret > 0) 940 ret = check_vcpu_requests(vcpu); 941 942 /* 943 * Preparing the interrupts to be injected also 944 * involves poking the GIC, which must be done in a 945 * non-preemptible context. 946 */ 947 preempt_disable(); 948 949 /* 950 * The VMID allocator only tracks active VMIDs per 951 * physical CPU, and therefore the VMID allocated may not be 952 * preserved on VMID roll-over if the task was preempted, 953 * making a thread's VMID inactive. So we need to call 954 * kvm_arm_vmid_update() in non-premptible context. 955 */ 956 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid); 957 958 kvm_pmu_flush_hwstate(vcpu); 959 960 local_irq_disable(); 961 962 kvm_vgic_flush_hwstate(vcpu); 963 964 kvm_pmu_update_vcpu_events(vcpu); 965 966 /* 967 * Ensure we set mode to IN_GUEST_MODE after we disable 968 * interrupts and before the final VCPU requests check. 969 * See the comment in kvm_vcpu_exiting_guest_mode() and 970 * Documentation/virt/kvm/vcpu-requests.rst 971 */ 972 smp_store_mb(vcpu->mode, IN_GUEST_MODE); 973 974 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) { 975 vcpu->mode = OUTSIDE_GUEST_MODE; 976 isb(); /* Ensure work in x_flush_hwstate is committed */ 977 kvm_pmu_sync_hwstate(vcpu); 978 if (static_branch_unlikely(&userspace_irqchip_in_use)) 979 kvm_timer_sync_user(vcpu); 980 kvm_vgic_sync_hwstate(vcpu); 981 local_irq_enable(); 982 preempt_enable(); 983 continue; 984 } 985 986 kvm_arm_setup_debug(vcpu); 987 kvm_arch_vcpu_ctxflush_fp(vcpu); 988 989 /************************************************************** 990 * Enter the guest 991 */ 992 trace_kvm_entry(*vcpu_pc(vcpu)); 993 guest_timing_enter_irqoff(); 994 995 ret = kvm_arm_vcpu_enter_exit(vcpu); 996 997 vcpu->mode = OUTSIDE_GUEST_MODE; 998 vcpu->stat.exits++; 999 /* 1000 * Back from guest 1001 *************************************************************/ 1002 1003 kvm_arm_clear_debug(vcpu); 1004 1005 /* 1006 * We must sync the PMU state before the vgic state so 1007 * that the vgic can properly sample the updated state of the 1008 * interrupt line. 1009 */ 1010 kvm_pmu_sync_hwstate(vcpu); 1011 1012 /* 1013 * Sync the vgic state before syncing the timer state because 1014 * the timer code needs to know if the virtual timer 1015 * interrupts are active. 1016 */ 1017 kvm_vgic_sync_hwstate(vcpu); 1018 1019 /* 1020 * Sync the timer hardware state before enabling interrupts as 1021 * we don't want vtimer interrupts to race with syncing the 1022 * timer virtual interrupt state. 1023 */ 1024 if (static_branch_unlikely(&userspace_irqchip_in_use)) 1025 kvm_timer_sync_user(vcpu); 1026 1027 kvm_arch_vcpu_ctxsync_fp(vcpu); 1028 1029 /* 1030 * We must ensure that any pending interrupts are taken before 1031 * we exit guest timing so that timer ticks are accounted as 1032 * guest time. Transiently unmask interrupts so that any 1033 * pending interrupts are taken. 1034 * 1035 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other 1036 * context synchronization event) is necessary to ensure that 1037 * pending interrupts are taken. 1038 */ 1039 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) { 1040 local_irq_enable(); 1041 isb(); 1042 local_irq_disable(); 1043 } 1044 1045 guest_timing_exit_irqoff(); 1046 1047 local_irq_enable(); 1048 1049 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); 1050 1051 /* Exit types that need handling before we can be preempted */ 1052 handle_exit_early(vcpu, ret); 1053 1054 preempt_enable(); 1055 1056 /* 1057 * The ARMv8 architecture doesn't give the hypervisor 1058 * a mechanism to prevent a guest from dropping to AArch32 EL0 1059 * if implemented by the CPU. If we spot the guest in such 1060 * state and that we decided it wasn't supposed to do so (like 1061 * with the asymmetric AArch32 case), return to userspace with 1062 * a fatal error. 1063 */ 1064 if (vcpu_mode_is_bad_32bit(vcpu)) { 1065 /* 1066 * As we have caught the guest red-handed, decide that 1067 * it isn't fit for purpose anymore by making the vcpu 1068 * invalid. The VMM can try and fix it by issuing a 1069 * KVM_ARM_VCPU_INIT if it really wants to. 1070 */ 1071 vcpu_clear_flag(vcpu, VCPU_INITIALIZED); 1072 ret = ARM_EXCEPTION_IL; 1073 } 1074 1075 ret = handle_exit(vcpu, ret); 1076 } 1077 1078 /* Tell userspace about in-kernel device output levels */ 1079 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { 1080 kvm_timer_update_run(vcpu); 1081 kvm_pmu_update_run(vcpu); 1082 } 1083 1084 kvm_sigset_deactivate(vcpu); 1085 1086 out: 1087 /* 1088 * In the unlikely event that we are returning to userspace 1089 * with pending exceptions or PC adjustment, commit these 1090 * adjustments in order to give userspace a consistent view of 1091 * the vcpu state. Note that this relies on __kvm_adjust_pc() 1092 * being preempt-safe on VHE. 1093 */ 1094 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) || 1095 vcpu_get_flag(vcpu, INCREMENT_PC))) 1096 kvm_call_hyp(__kvm_adjust_pc, vcpu); 1097 1098 vcpu_put(vcpu); 1099 return ret; 1100 } 1101 1102 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) 1103 { 1104 int bit_index; 1105 bool set; 1106 unsigned long *hcr; 1107 1108 if (number == KVM_ARM_IRQ_CPU_IRQ) 1109 bit_index = __ffs(HCR_VI); 1110 else /* KVM_ARM_IRQ_CPU_FIQ */ 1111 bit_index = __ffs(HCR_VF); 1112 1113 hcr = vcpu_hcr(vcpu); 1114 if (level) 1115 set = test_and_set_bit(bit_index, hcr); 1116 else 1117 set = test_and_clear_bit(bit_index, hcr); 1118 1119 /* 1120 * If we didn't change anything, no need to wake up or kick other CPUs 1121 */ 1122 if (set == level) 1123 return 0; 1124 1125 /* 1126 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and 1127 * trigger a world-switch round on the running physical CPU to set the 1128 * virtual IRQ/FIQ fields in the HCR appropriately. 1129 */ 1130 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); 1131 kvm_vcpu_kick(vcpu); 1132 1133 return 0; 1134 } 1135 1136 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 1137 bool line_status) 1138 { 1139 u32 irq = irq_level->irq; 1140 unsigned int irq_type, vcpu_idx, irq_num; 1141 int nrcpus = atomic_read(&kvm->online_vcpus); 1142 struct kvm_vcpu *vcpu = NULL; 1143 bool level = irq_level->level; 1144 1145 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; 1146 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; 1147 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); 1148 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; 1149 1150 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); 1151 1152 switch (irq_type) { 1153 case KVM_ARM_IRQ_TYPE_CPU: 1154 if (irqchip_in_kernel(kvm)) 1155 return -ENXIO; 1156 1157 if (vcpu_idx >= nrcpus) 1158 return -EINVAL; 1159 1160 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 1161 if (!vcpu) 1162 return -EINVAL; 1163 1164 if (irq_num > KVM_ARM_IRQ_CPU_FIQ) 1165 return -EINVAL; 1166 1167 return vcpu_interrupt_line(vcpu, irq_num, level); 1168 case KVM_ARM_IRQ_TYPE_PPI: 1169 if (!irqchip_in_kernel(kvm)) 1170 return -ENXIO; 1171 1172 if (vcpu_idx >= nrcpus) 1173 return -EINVAL; 1174 1175 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 1176 if (!vcpu) 1177 return -EINVAL; 1178 1179 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) 1180 return -EINVAL; 1181 1182 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); 1183 case KVM_ARM_IRQ_TYPE_SPI: 1184 if (!irqchip_in_kernel(kvm)) 1185 return -ENXIO; 1186 1187 if (irq_num < VGIC_NR_PRIVATE_IRQS) 1188 return -EINVAL; 1189 1190 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); 1191 } 1192 1193 return -EINVAL; 1194 } 1195 1196 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu, 1197 const struct kvm_vcpu_init *init) 1198 { 1199 unsigned long features = init->features[0]; 1200 int i; 1201 1202 if (features & ~KVM_VCPU_VALID_FEATURES) 1203 return -ENOENT; 1204 1205 for (i = 1; i < ARRAY_SIZE(init->features); i++) { 1206 if (init->features[i]) 1207 return -ENOENT; 1208 } 1209 1210 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features)) 1211 return 0; 1212 1213 if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1)) 1214 return -EINVAL; 1215 1216 /* MTE is incompatible with AArch32 */ 1217 if (kvm_has_mte(vcpu->kvm)) 1218 return -EINVAL; 1219 1220 /* NV is incompatible with AArch32 */ 1221 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features)) 1222 return -EINVAL; 1223 1224 return 0; 1225 } 1226 1227 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu, 1228 const struct kvm_vcpu_init *init) 1229 { 1230 unsigned long features = init->features[0]; 1231 1232 return !bitmap_equal(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES); 1233 } 1234 1235 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 1236 const struct kvm_vcpu_init *init) 1237 { 1238 unsigned long features = init->features[0]; 1239 struct kvm *kvm = vcpu->kvm; 1240 int ret = -EINVAL; 1241 1242 mutex_lock(&kvm->arch.config_lock); 1243 1244 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) && 1245 !bitmap_equal(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES)) 1246 goto out_unlock; 1247 1248 bitmap_copy(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES); 1249 1250 /* Now we know what it is, we can reset it. */ 1251 ret = kvm_reset_vcpu(vcpu); 1252 if (ret) { 1253 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 1254 goto out_unlock; 1255 } 1256 1257 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES); 1258 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags); 1259 vcpu_set_flag(vcpu, VCPU_INITIALIZED); 1260 out_unlock: 1261 mutex_unlock(&kvm->arch.config_lock); 1262 return ret; 1263 } 1264 1265 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 1266 const struct kvm_vcpu_init *init) 1267 { 1268 int ret; 1269 1270 if (init->target != KVM_ARM_TARGET_GENERIC_V8 && 1271 init->target != kvm_target_cpu()) 1272 return -EINVAL; 1273 1274 ret = kvm_vcpu_init_check_features(vcpu, init); 1275 if (ret) 1276 return ret; 1277 1278 if (!kvm_vcpu_initialized(vcpu)) 1279 return __kvm_vcpu_set_target(vcpu, init); 1280 1281 if (kvm_vcpu_init_changed(vcpu, init)) 1282 return -EINVAL; 1283 1284 return kvm_reset_vcpu(vcpu); 1285 } 1286 1287 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, 1288 struct kvm_vcpu_init *init) 1289 { 1290 bool power_off = false; 1291 int ret; 1292 1293 /* 1294 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid 1295 * reflecting it in the finalized feature set, thus limiting its scope 1296 * to a single KVM_ARM_VCPU_INIT call. 1297 */ 1298 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) { 1299 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF); 1300 power_off = true; 1301 } 1302 1303 ret = kvm_vcpu_set_target(vcpu, init); 1304 if (ret) 1305 return ret; 1306 1307 /* 1308 * Ensure a rebooted VM will fault in RAM pages and detect if the 1309 * guest MMU is turned off and flush the caches as needed. 1310 * 1311 * S2FWB enforces all memory accesses to RAM being cacheable, 1312 * ensuring that the data side is always coherent. We still 1313 * need to invalidate the I-cache though, as FWB does *not* 1314 * imply CTR_EL0.DIC. 1315 */ 1316 if (vcpu_has_run_once(vcpu)) { 1317 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB)) 1318 stage2_unmap_vm(vcpu->kvm); 1319 else 1320 icache_inval_all_pou(); 1321 } 1322 1323 vcpu_reset_hcr(vcpu); 1324 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu); 1325 1326 /* 1327 * Handle the "start in power-off" case. 1328 */ 1329 spin_lock(&vcpu->arch.mp_state_lock); 1330 1331 if (power_off) 1332 __kvm_arm_vcpu_power_off(vcpu); 1333 else 1334 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE); 1335 1336 spin_unlock(&vcpu->arch.mp_state_lock); 1337 1338 return 0; 1339 } 1340 1341 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, 1342 struct kvm_device_attr *attr) 1343 { 1344 int ret = -ENXIO; 1345 1346 switch (attr->group) { 1347 default: 1348 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); 1349 break; 1350 } 1351 1352 return ret; 1353 } 1354 1355 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, 1356 struct kvm_device_attr *attr) 1357 { 1358 int ret = -ENXIO; 1359 1360 switch (attr->group) { 1361 default: 1362 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); 1363 break; 1364 } 1365 1366 return ret; 1367 } 1368 1369 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, 1370 struct kvm_device_attr *attr) 1371 { 1372 int ret = -ENXIO; 1373 1374 switch (attr->group) { 1375 default: 1376 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); 1377 break; 1378 } 1379 1380 return ret; 1381 } 1382 1383 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1384 struct kvm_vcpu_events *events) 1385 { 1386 memset(events, 0, sizeof(*events)); 1387 1388 return __kvm_arm_vcpu_get_events(vcpu, events); 1389 } 1390 1391 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1392 struct kvm_vcpu_events *events) 1393 { 1394 int i; 1395 1396 /* check whether the reserved field is zero */ 1397 for (i = 0; i < ARRAY_SIZE(events->reserved); i++) 1398 if (events->reserved[i]) 1399 return -EINVAL; 1400 1401 /* check whether the pad field is zero */ 1402 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) 1403 if (events->exception.pad[i]) 1404 return -EINVAL; 1405 1406 return __kvm_arm_vcpu_set_events(vcpu, events); 1407 } 1408 1409 long kvm_arch_vcpu_ioctl(struct file *filp, 1410 unsigned int ioctl, unsigned long arg) 1411 { 1412 struct kvm_vcpu *vcpu = filp->private_data; 1413 void __user *argp = (void __user *)arg; 1414 struct kvm_device_attr attr; 1415 long r; 1416 1417 switch (ioctl) { 1418 case KVM_ARM_VCPU_INIT: { 1419 struct kvm_vcpu_init init; 1420 1421 r = -EFAULT; 1422 if (copy_from_user(&init, argp, sizeof(init))) 1423 break; 1424 1425 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); 1426 break; 1427 } 1428 case KVM_SET_ONE_REG: 1429 case KVM_GET_ONE_REG: { 1430 struct kvm_one_reg reg; 1431 1432 r = -ENOEXEC; 1433 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1434 break; 1435 1436 r = -EFAULT; 1437 if (copy_from_user(®, argp, sizeof(reg))) 1438 break; 1439 1440 /* 1441 * We could owe a reset due to PSCI. Handle the pending reset 1442 * here to ensure userspace register accesses are ordered after 1443 * the reset. 1444 */ 1445 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 1446 kvm_reset_vcpu(vcpu); 1447 1448 if (ioctl == KVM_SET_ONE_REG) 1449 r = kvm_arm_set_reg(vcpu, ®); 1450 else 1451 r = kvm_arm_get_reg(vcpu, ®); 1452 break; 1453 } 1454 case KVM_GET_REG_LIST: { 1455 struct kvm_reg_list __user *user_list = argp; 1456 struct kvm_reg_list reg_list; 1457 unsigned n; 1458 1459 r = -ENOEXEC; 1460 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1461 break; 1462 1463 r = -EPERM; 1464 if (!kvm_arm_vcpu_is_finalized(vcpu)) 1465 break; 1466 1467 r = -EFAULT; 1468 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 1469 break; 1470 n = reg_list.n; 1471 reg_list.n = kvm_arm_num_regs(vcpu); 1472 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 1473 break; 1474 r = -E2BIG; 1475 if (n < reg_list.n) 1476 break; 1477 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); 1478 break; 1479 } 1480 case KVM_SET_DEVICE_ATTR: { 1481 r = -EFAULT; 1482 if (copy_from_user(&attr, argp, sizeof(attr))) 1483 break; 1484 r = kvm_arm_vcpu_set_attr(vcpu, &attr); 1485 break; 1486 } 1487 case KVM_GET_DEVICE_ATTR: { 1488 r = -EFAULT; 1489 if (copy_from_user(&attr, argp, sizeof(attr))) 1490 break; 1491 r = kvm_arm_vcpu_get_attr(vcpu, &attr); 1492 break; 1493 } 1494 case KVM_HAS_DEVICE_ATTR: { 1495 r = -EFAULT; 1496 if (copy_from_user(&attr, argp, sizeof(attr))) 1497 break; 1498 r = kvm_arm_vcpu_has_attr(vcpu, &attr); 1499 break; 1500 } 1501 case KVM_GET_VCPU_EVENTS: { 1502 struct kvm_vcpu_events events; 1503 1504 if (kvm_arm_vcpu_get_events(vcpu, &events)) 1505 return -EINVAL; 1506 1507 if (copy_to_user(argp, &events, sizeof(events))) 1508 return -EFAULT; 1509 1510 return 0; 1511 } 1512 case KVM_SET_VCPU_EVENTS: { 1513 struct kvm_vcpu_events events; 1514 1515 if (copy_from_user(&events, argp, sizeof(events))) 1516 return -EFAULT; 1517 1518 return kvm_arm_vcpu_set_events(vcpu, &events); 1519 } 1520 case KVM_ARM_VCPU_FINALIZE: { 1521 int what; 1522 1523 if (!kvm_vcpu_initialized(vcpu)) 1524 return -ENOEXEC; 1525 1526 if (get_user(what, (const int __user *)argp)) 1527 return -EFAULT; 1528 1529 return kvm_arm_vcpu_finalize(vcpu, what); 1530 } 1531 default: 1532 r = -EINVAL; 1533 } 1534 1535 return r; 1536 } 1537 1538 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 1539 { 1540 1541 } 1542 1543 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, 1544 struct kvm_arm_device_addr *dev_addr) 1545 { 1546 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) { 1547 case KVM_ARM_DEVICE_VGIC_V2: 1548 if (!vgic_present) 1549 return -ENXIO; 1550 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr); 1551 default: 1552 return -ENODEV; 1553 } 1554 } 1555 1556 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr) 1557 { 1558 switch (attr->group) { 1559 case KVM_ARM_VM_SMCCC_CTRL: 1560 return kvm_vm_smccc_has_attr(kvm, attr); 1561 default: 1562 return -ENXIO; 1563 } 1564 } 1565 1566 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr) 1567 { 1568 switch (attr->group) { 1569 case KVM_ARM_VM_SMCCC_CTRL: 1570 return kvm_vm_smccc_set_attr(kvm, attr); 1571 default: 1572 return -ENXIO; 1573 } 1574 } 1575 1576 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg) 1577 { 1578 struct kvm *kvm = filp->private_data; 1579 void __user *argp = (void __user *)arg; 1580 struct kvm_device_attr attr; 1581 1582 switch (ioctl) { 1583 case KVM_CREATE_IRQCHIP: { 1584 int ret; 1585 if (!vgic_present) 1586 return -ENXIO; 1587 mutex_lock(&kvm->lock); 1588 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); 1589 mutex_unlock(&kvm->lock); 1590 return ret; 1591 } 1592 case KVM_ARM_SET_DEVICE_ADDR: { 1593 struct kvm_arm_device_addr dev_addr; 1594 1595 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) 1596 return -EFAULT; 1597 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); 1598 } 1599 case KVM_ARM_PREFERRED_TARGET: { 1600 struct kvm_vcpu_init init = { 1601 .target = KVM_ARM_TARGET_GENERIC_V8, 1602 }; 1603 1604 if (copy_to_user(argp, &init, sizeof(init))) 1605 return -EFAULT; 1606 1607 return 0; 1608 } 1609 case KVM_ARM_MTE_COPY_TAGS: { 1610 struct kvm_arm_copy_mte_tags copy_tags; 1611 1612 if (copy_from_user(©_tags, argp, sizeof(copy_tags))) 1613 return -EFAULT; 1614 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags); 1615 } 1616 case KVM_ARM_SET_COUNTER_OFFSET: { 1617 struct kvm_arm_counter_offset offset; 1618 1619 if (copy_from_user(&offset, argp, sizeof(offset))) 1620 return -EFAULT; 1621 return kvm_vm_ioctl_set_counter_offset(kvm, &offset); 1622 } 1623 case KVM_HAS_DEVICE_ATTR: { 1624 if (copy_from_user(&attr, argp, sizeof(attr))) 1625 return -EFAULT; 1626 1627 return kvm_vm_has_attr(kvm, &attr); 1628 } 1629 case KVM_SET_DEVICE_ATTR: { 1630 if (copy_from_user(&attr, argp, sizeof(attr))) 1631 return -EFAULT; 1632 1633 return kvm_vm_set_attr(kvm, &attr); 1634 } 1635 default: 1636 return -EINVAL; 1637 } 1638 } 1639 1640 /* unlocks vcpus from @vcpu_lock_idx and smaller */ 1641 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx) 1642 { 1643 struct kvm_vcpu *tmp_vcpu; 1644 1645 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) { 1646 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx); 1647 mutex_unlock(&tmp_vcpu->mutex); 1648 } 1649 } 1650 1651 void unlock_all_vcpus(struct kvm *kvm) 1652 { 1653 lockdep_assert_held(&kvm->lock); 1654 1655 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1); 1656 } 1657 1658 /* Returns true if all vcpus were locked, false otherwise */ 1659 bool lock_all_vcpus(struct kvm *kvm) 1660 { 1661 struct kvm_vcpu *tmp_vcpu; 1662 unsigned long c; 1663 1664 lockdep_assert_held(&kvm->lock); 1665 1666 /* 1667 * Any time a vcpu is in an ioctl (including running), the 1668 * core KVM code tries to grab the vcpu->mutex. 1669 * 1670 * By grabbing the vcpu->mutex of all VCPUs we ensure that no 1671 * other VCPUs can fiddle with the state while we access it. 1672 */ 1673 kvm_for_each_vcpu(c, tmp_vcpu, kvm) { 1674 if (!mutex_trylock(&tmp_vcpu->mutex)) { 1675 unlock_vcpus(kvm, c - 1); 1676 return false; 1677 } 1678 } 1679 1680 return true; 1681 } 1682 1683 static unsigned long nvhe_percpu_size(void) 1684 { 1685 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) - 1686 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start); 1687 } 1688 1689 static unsigned long nvhe_percpu_order(void) 1690 { 1691 unsigned long size = nvhe_percpu_size(); 1692 1693 return size ? get_order(size) : 0; 1694 } 1695 1696 /* A lookup table holding the hypervisor VA for each vector slot */ 1697 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS]; 1698 1699 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot) 1700 { 1701 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot); 1702 } 1703 1704 static int kvm_init_vector_slots(void) 1705 { 1706 int err; 1707 void *base; 1708 1709 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector)); 1710 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT); 1711 1712 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs)); 1713 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT); 1714 1715 if (kvm_system_needs_idmapped_vectors() && 1716 !is_protected_kvm_enabled()) { 1717 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs), 1718 __BP_HARDEN_HYP_VECS_SZ, &base); 1719 if (err) 1720 return err; 1721 } 1722 1723 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT); 1724 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT); 1725 return 0; 1726 } 1727 1728 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits) 1729 { 1730 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu); 1731 unsigned long tcr; 1732 1733 /* 1734 * Calculate the raw per-cpu offset without a translation from the 1735 * kernel's mapping to the linear mapping, and store it in tpidr_el2 1736 * so that we can use adr_l to access per-cpu variables in EL2. 1737 * Also drop the KASAN tag which gets in the way... 1738 */ 1739 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) - 1740 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start)); 1741 1742 params->mair_el2 = read_sysreg(mair_el1); 1743 1744 tcr = read_sysreg(tcr_el1); 1745 if (cpus_have_final_cap(ARM64_KVM_HVHE)) { 1746 tcr |= TCR_EPD1_MASK; 1747 } else { 1748 tcr &= TCR_EL2_MASK; 1749 tcr |= TCR_EL2_RES1; 1750 } 1751 tcr &= ~TCR_T0SZ_MASK; 1752 tcr |= TCR_T0SZ(hyp_va_bits); 1753 params->tcr_el2 = tcr; 1754 1755 params->pgd_pa = kvm_mmu_get_httbr(); 1756 if (is_protected_kvm_enabled()) 1757 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS; 1758 else 1759 params->hcr_el2 = HCR_HOST_NVHE_FLAGS; 1760 if (cpus_have_final_cap(ARM64_KVM_HVHE)) 1761 params->hcr_el2 |= HCR_E2H; 1762 params->vttbr = params->vtcr = 0; 1763 1764 /* 1765 * Flush the init params from the data cache because the struct will 1766 * be read while the MMU is off. 1767 */ 1768 kvm_flush_dcache_to_poc(params, sizeof(*params)); 1769 } 1770 1771 static void hyp_install_host_vector(void) 1772 { 1773 struct kvm_nvhe_init_params *params; 1774 struct arm_smccc_res res; 1775 1776 /* Switch from the HYP stub to our own HYP init vector */ 1777 __hyp_set_vectors(kvm_get_idmap_vector()); 1778 1779 /* 1780 * Call initialization code, and switch to the full blown HYP code. 1781 * If the cpucaps haven't been finalized yet, something has gone very 1782 * wrong, and hyp will crash and burn when it uses any 1783 * cpus_have_const_cap() wrapper. 1784 */ 1785 BUG_ON(!system_capabilities_finalized()); 1786 params = this_cpu_ptr_nvhe_sym(kvm_init_params); 1787 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res); 1788 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); 1789 } 1790 1791 static void cpu_init_hyp_mode(void) 1792 { 1793 hyp_install_host_vector(); 1794 1795 /* 1796 * Disabling SSBD on a non-VHE system requires us to enable SSBS 1797 * at EL2. 1798 */ 1799 if (this_cpu_has_cap(ARM64_SSBS) && 1800 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) { 1801 kvm_call_hyp_nvhe(__kvm_enable_ssbs); 1802 } 1803 } 1804 1805 static void cpu_hyp_reset(void) 1806 { 1807 if (!is_kernel_in_hyp_mode()) 1808 __hyp_reset_vectors(); 1809 } 1810 1811 /* 1812 * EL2 vectors can be mapped and rerouted in a number of ways, 1813 * depending on the kernel configuration and CPU present: 1814 * 1815 * - If the CPU is affected by Spectre-v2, the hardening sequence is 1816 * placed in one of the vector slots, which is executed before jumping 1817 * to the real vectors. 1818 * 1819 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot 1820 * containing the hardening sequence is mapped next to the idmap page, 1821 * and executed before jumping to the real vectors. 1822 * 1823 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an 1824 * empty slot is selected, mapped next to the idmap page, and 1825 * executed before jumping to the real vectors. 1826 * 1827 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with 1828 * VHE, as we don't have hypervisor-specific mappings. If the system 1829 * is VHE and yet selects this capability, it will be ignored. 1830 */ 1831 static void cpu_set_hyp_vector(void) 1832 { 1833 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data); 1834 void *vector = hyp_spectre_vector_selector[data->slot]; 1835 1836 if (!is_protected_kvm_enabled()) 1837 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector; 1838 else 1839 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot); 1840 } 1841 1842 static void cpu_hyp_init_context(void) 1843 { 1844 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt); 1845 1846 if (!is_kernel_in_hyp_mode()) 1847 cpu_init_hyp_mode(); 1848 } 1849 1850 static void cpu_hyp_init_features(void) 1851 { 1852 cpu_set_hyp_vector(); 1853 kvm_arm_init_debug(); 1854 1855 if (is_kernel_in_hyp_mode()) 1856 kvm_timer_init_vhe(); 1857 1858 if (vgic_present) 1859 kvm_vgic_init_cpu_hardware(); 1860 } 1861 1862 static void cpu_hyp_reinit(void) 1863 { 1864 cpu_hyp_reset(); 1865 cpu_hyp_init_context(); 1866 cpu_hyp_init_features(); 1867 } 1868 1869 static void cpu_hyp_init(void *discard) 1870 { 1871 if (!__this_cpu_read(kvm_hyp_initialized)) { 1872 cpu_hyp_reinit(); 1873 __this_cpu_write(kvm_hyp_initialized, 1); 1874 } 1875 } 1876 1877 static void cpu_hyp_uninit(void *discard) 1878 { 1879 if (__this_cpu_read(kvm_hyp_initialized)) { 1880 cpu_hyp_reset(); 1881 __this_cpu_write(kvm_hyp_initialized, 0); 1882 } 1883 } 1884 1885 int kvm_arch_hardware_enable(void) 1886 { 1887 /* 1888 * Most calls to this function are made with migration 1889 * disabled, but not with preemption disabled. The former is 1890 * enough to ensure correctness, but most of the helpers 1891 * expect the later and will throw a tantrum otherwise. 1892 */ 1893 preempt_disable(); 1894 1895 cpu_hyp_init(NULL); 1896 1897 kvm_vgic_cpu_up(); 1898 kvm_timer_cpu_up(); 1899 1900 preempt_enable(); 1901 1902 return 0; 1903 } 1904 1905 void kvm_arch_hardware_disable(void) 1906 { 1907 kvm_timer_cpu_down(); 1908 kvm_vgic_cpu_down(); 1909 1910 if (!is_protected_kvm_enabled()) 1911 cpu_hyp_uninit(NULL); 1912 } 1913 1914 #ifdef CONFIG_CPU_PM 1915 static int hyp_init_cpu_pm_notifier(struct notifier_block *self, 1916 unsigned long cmd, 1917 void *v) 1918 { 1919 /* 1920 * kvm_hyp_initialized is left with its old value over 1921 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should 1922 * re-enable hyp. 1923 */ 1924 switch (cmd) { 1925 case CPU_PM_ENTER: 1926 if (__this_cpu_read(kvm_hyp_initialized)) 1927 /* 1928 * don't update kvm_hyp_initialized here 1929 * so that the hyp will be re-enabled 1930 * when we resume. See below. 1931 */ 1932 cpu_hyp_reset(); 1933 1934 return NOTIFY_OK; 1935 case CPU_PM_ENTER_FAILED: 1936 case CPU_PM_EXIT: 1937 if (__this_cpu_read(kvm_hyp_initialized)) 1938 /* The hyp was enabled before suspend. */ 1939 cpu_hyp_reinit(); 1940 1941 return NOTIFY_OK; 1942 1943 default: 1944 return NOTIFY_DONE; 1945 } 1946 } 1947 1948 static struct notifier_block hyp_init_cpu_pm_nb = { 1949 .notifier_call = hyp_init_cpu_pm_notifier, 1950 }; 1951 1952 static void __init hyp_cpu_pm_init(void) 1953 { 1954 if (!is_protected_kvm_enabled()) 1955 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); 1956 } 1957 static void __init hyp_cpu_pm_exit(void) 1958 { 1959 if (!is_protected_kvm_enabled()) 1960 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); 1961 } 1962 #else 1963 static inline void __init hyp_cpu_pm_init(void) 1964 { 1965 } 1966 static inline void __init hyp_cpu_pm_exit(void) 1967 { 1968 } 1969 #endif 1970 1971 static void __init init_cpu_logical_map(void) 1972 { 1973 unsigned int cpu; 1974 1975 /* 1976 * Copy the MPIDR <-> logical CPU ID mapping to hyp. 1977 * Only copy the set of online CPUs whose features have been checked 1978 * against the finalized system capabilities. The hypervisor will not 1979 * allow any other CPUs from the `possible` set to boot. 1980 */ 1981 for_each_online_cpu(cpu) 1982 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu); 1983 } 1984 1985 #define init_psci_0_1_impl_state(config, what) \ 1986 config.psci_0_1_ ## what ## _implemented = psci_ops.what 1987 1988 static bool __init init_psci_relay(void) 1989 { 1990 /* 1991 * If PSCI has not been initialized, protected KVM cannot install 1992 * itself on newly booted CPUs. 1993 */ 1994 if (!psci_ops.get_version) { 1995 kvm_err("Cannot initialize protected mode without PSCI\n"); 1996 return false; 1997 } 1998 1999 kvm_host_psci_config.version = psci_ops.get_version(); 2000 kvm_host_psci_config.smccc_version = arm_smccc_get_version(); 2001 2002 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) { 2003 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids(); 2004 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend); 2005 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on); 2006 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off); 2007 init_psci_0_1_impl_state(kvm_host_psci_config, migrate); 2008 } 2009 return true; 2010 } 2011 2012 static int __init init_subsystems(void) 2013 { 2014 int err = 0; 2015 2016 /* 2017 * Enable hardware so that subsystem initialisation can access EL2. 2018 */ 2019 on_each_cpu(cpu_hyp_init, NULL, 1); 2020 2021 /* 2022 * Register CPU lower-power notifier 2023 */ 2024 hyp_cpu_pm_init(); 2025 2026 /* 2027 * Init HYP view of VGIC 2028 */ 2029 err = kvm_vgic_hyp_init(); 2030 switch (err) { 2031 case 0: 2032 vgic_present = true; 2033 break; 2034 case -ENODEV: 2035 case -ENXIO: 2036 vgic_present = false; 2037 err = 0; 2038 break; 2039 default: 2040 goto out; 2041 } 2042 2043 /* 2044 * Init HYP architected timer support 2045 */ 2046 err = kvm_timer_hyp_init(vgic_present); 2047 if (err) 2048 goto out; 2049 2050 kvm_register_perf_callbacks(NULL); 2051 2052 out: 2053 if (err) 2054 hyp_cpu_pm_exit(); 2055 2056 if (err || !is_protected_kvm_enabled()) 2057 on_each_cpu(cpu_hyp_uninit, NULL, 1); 2058 2059 return err; 2060 } 2061 2062 static void __init teardown_subsystems(void) 2063 { 2064 kvm_unregister_perf_callbacks(); 2065 hyp_cpu_pm_exit(); 2066 } 2067 2068 static void __init teardown_hyp_mode(void) 2069 { 2070 int cpu; 2071 2072 free_hyp_pgds(); 2073 for_each_possible_cpu(cpu) { 2074 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); 2075 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order()); 2076 } 2077 } 2078 2079 static int __init do_pkvm_init(u32 hyp_va_bits) 2080 { 2081 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)); 2082 int ret; 2083 2084 preempt_disable(); 2085 cpu_hyp_init_context(); 2086 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size, 2087 num_possible_cpus(), kern_hyp_va(per_cpu_base), 2088 hyp_va_bits); 2089 cpu_hyp_init_features(); 2090 2091 /* 2092 * The stub hypercalls are now disabled, so set our local flag to 2093 * prevent a later re-init attempt in kvm_arch_hardware_enable(). 2094 */ 2095 __this_cpu_write(kvm_hyp_initialized, 1); 2096 preempt_enable(); 2097 2098 return ret; 2099 } 2100 2101 static u64 get_hyp_id_aa64pfr0_el1(void) 2102 { 2103 /* 2104 * Track whether the system isn't affected by spectre/meltdown in the 2105 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs. 2106 * Although this is per-CPU, we make it global for simplicity, e.g., not 2107 * to have to worry about vcpu migration. 2108 * 2109 * Unlike for non-protected VMs, userspace cannot override this for 2110 * protected VMs. 2111 */ 2112 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1); 2113 2114 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) | 2115 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3)); 2116 2117 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2), 2118 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED); 2119 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3), 2120 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED); 2121 2122 return val; 2123 } 2124 2125 static void kvm_hyp_init_symbols(void) 2126 { 2127 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1(); 2128 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1); 2129 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1); 2130 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1); 2131 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1); 2132 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1); 2133 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); 2134 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1); 2135 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1); 2136 kvm_nvhe_sym(__icache_flags) = __icache_flags; 2137 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits; 2138 } 2139 2140 static int __init kvm_hyp_init_protection(u32 hyp_va_bits) 2141 { 2142 void *addr = phys_to_virt(hyp_mem_base); 2143 int ret; 2144 2145 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP); 2146 if (ret) 2147 return ret; 2148 2149 ret = do_pkvm_init(hyp_va_bits); 2150 if (ret) 2151 return ret; 2152 2153 free_hyp_pgds(); 2154 2155 return 0; 2156 } 2157 2158 static void pkvm_hyp_init_ptrauth(void) 2159 { 2160 struct kvm_cpu_context *hyp_ctxt; 2161 int cpu; 2162 2163 for_each_possible_cpu(cpu) { 2164 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu); 2165 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long(); 2166 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long(); 2167 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long(); 2168 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long(); 2169 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long(); 2170 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long(); 2171 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long(); 2172 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long(); 2173 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long(); 2174 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long(); 2175 } 2176 } 2177 2178 /* Inits Hyp-mode on all online CPUs */ 2179 static int __init init_hyp_mode(void) 2180 { 2181 u32 hyp_va_bits; 2182 int cpu; 2183 int err = -ENOMEM; 2184 2185 /* 2186 * The protected Hyp-mode cannot be initialized if the memory pool 2187 * allocation has failed. 2188 */ 2189 if (is_protected_kvm_enabled() && !hyp_mem_base) 2190 goto out_err; 2191 2192 /* 2193 * Allocate Hyp PGD and setup Hyp identity mapping 2194 */ 2195 err = kvm_mmu_init(&hyp_va_bits); 2196 if (err) 2197 goto out_err; 2198 2199 /* 2200 * Allocate stack pages for Hypervisor-mode 2201 */ 2202 for_each_possible_cpu(cpu) { 2203 unsigned long stack_page; 2204 2205 stack_page = __get_free_page(GFP_KERNEL); 2206 if (!stack_page) { 2207 err = -ENOMEM; 2208 goto out_err; 2209 } 2210 2211 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; 2212 } 2213 2214 /* 2215 * Allocate and initialize pages for Hypervisor-mode percpu regions. 2216 */ 2217 for_each_possible_cpu(cpu) { 2218 struct page *page; 2219 void *page_addr; 2220 2221 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order()); 2222 if (!page) { 2223 err = -ENOMEM; 2224 goto out_err; 2225 } 2226 2227 page_addr = page_address(page); 2228 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size()); 2229 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr; 2230 } 2231 2232 /* 2233 * Map the Hyp-code called directly from the host 2234 */ 2235 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), 2236 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); 2237 if (err) { 2238 kvm_err("Cannot map world-switch code\n"); 2239 goto out_err; 2240 } 2241 2242 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start), 2243 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO); 2244 if (err) { 2245 kvm_err("Cannot map .hyp.rodata section\n"); 2246 goto out_err; 2247 } 2248 2249 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), 2250 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); 2251 if (err) { 2252 kvm_err("Cannot map rodata section\n"); 2253 goto out_err; 2254 } 2255 2256 /* 2257 * .hyp.bss is guaranteed to be placed at the beginning of the .bss 2258 * section thanks to an assertion in the linker script. Map it RW and 2259 * the rest of .bss RO. 2260 */ 2261 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start), 2262 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP); 2263 if (err) { 2264 kvm_err("Cannot map hyp bss section: %d\n", err); 2265 goto out_err; 2266 } 2267 2268 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end), 2269 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); 2270 if (err) { 2271 kvm_err("Cannot map bss section\n"); 2272 goto out_err; 2273 } 2274 2275 /* 2276 * Map the Hyp stack pages 2277 */ 2278 for_each_possible_cpu(cpu) { 2279 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu); 2280 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); 2281 2282 err = create_hyp_stack(__pa(stack_page), ¶ms->stack_hyp_va); 2283 if (err) { 2284 kvm_err("Cannot map hyp stack\n"); 2285 goto out_err; 2286 } 2287 2288 /* 2289 * Save the stack PA in nvhe_init_params. This will be needed 2290 * to recreate the stack mapping in protected nVHE mode. 2291 * __hyp_pa() won't do the right thing there, since the stack 2292 * has been mapped in the flexible private VA space. 2293 */ 2294 params->stack_pa = __pa(stack_page); 2295 } 2296 2297 for_each_possible_cpu(cpu) { 2298 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu]; 2299 char *percpu_end = percpu_begin + nvhe_percpu_size(); 2300 2301 /* Map Hyp percpu pages */ 2302 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP); 2303 if (err) { 2304 kvm_err("Cannot map hyp percpu region\n"); 2305 goto out_err; 2306 } 2307 2308 /* Prepare the CPU initialization parameters */ 2309 cpu_prepare_hyp_mode(cpu, hyp_va_bits); 2310 } 2311 2312 kvm_hyp_init_symbols(); 2313 2314 if (is_protected_kvm_enabled()) { 2315 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) && 2316 cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH)) 2317 pkvm_hyp_init_ptrauth(); 2318 2319 init_cpu_logical_map(); 2320 2321 if (!init_psci_relay()) { 2322 err = -ENODEV; 2323 goto out_err; 2324 } 2325 2326 err = kvm_hyp_init_protection(hyp_va_bits); 2327 if (err) { 2328 kvm_err("Failed to init hyp memory protection\n"); 2329 goto out_err; 2330 } 2331 } 2332 2333 return 0; 2334 2335 out_err: 2336 teardown_hyp_mode(); 2337 kvm_err("error initializing Hyp mode: %d\n", err); 2338 return err; 2339 } 2340 2341 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) 2342 { 2343 struct kvm_vcpu *vcpu; 2344 unsigned long i; 2345 2346 mpidr &= MPIDR_HWID_BITMASK; 2347 kvm_for_each_vcpu(i, vcpu, kvm) { 2348 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) 2349 return vcpu; 2350 } 2351 return NULL; 2352 } 2353 2354 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm) 2355 { 2356 return irqchip_in_kernel(kvm); 2357 } 2358 2359 bool kvm_arch_has_irq_bypass(void) 2360 { 2361 return true; 2362 } 2363 2364 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 2365 struct irq_bypass_producer *prod) 2366 { 2367 struct kvm_kernel_irqfd *irqfd = 2368 container_of(cons, struct kvm_kernel_irqfd, consumer); 2369 2370 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, 2371 &irqfd->irq_entry); 2372 } 2373 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 2374 struct irq_bypass_producer *prod) 2375 { 2376 struct kvm_kernel_irqfd *irqfd = 2377 container_of(cons, struct kvm_kernel_irqfd, consumer); 2378 2379 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, 2380 &irqfd->irq_entry); 2381 } 2382 2383 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) 2384 { 2385 struct kvm_kernel_irqfd *irqfd = 2386 container_of(cons, struct kvm_kernel_irqfd, consumer); 2387 2388 kvm_arm_halt_guest(irqfd->kvm); 2389 } 2390 2391 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) 2392 { 2393 struct kvm_kernel_irqfd *irqfd = 2394 container_of(cons, struct kvm_kernel_irqfd, consumer); 2395 2396 kvm_arm_resume_guest(irqfd->kvm); 2397 } 2398 2399 /* Initialize Hyp-mode and memory mappings on all CPUs */ 2400 static __init int kvm_arm_init(void) 2401 { 2402 int err; 2403 bool in_hyp_mode; 2404 2405 if (!is_hyp_mode_available()) { 2406 kvm_info("HYP mode not available\n"); 2407 return -ENODEV; 2408 } 2409 2410 if (kvm_get_mode() == KVM_MODE_NONE) { 2411 kvm_info("KVM disabled from command line\n"); 2412 return -ENODEV; 2413 } 2414 2415 err = kvm_sys_reg_table_init(); 2416 if (err) { 2417 kvm_info("Error initializing system register tables"); 2418 return err; 2419 } 2420 2421 in_hyp_mode = is_kernel_in_hyp_mode(); 2422 2423 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) || 2424 cpus_have_final_cap(ARM64_WORKAROUND_1508412)) 2425 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \ 2426 "Only trusted guests should be used on this system.\n"); 2427 2428 err = kvm_set_ipa_limit(); 2429 if (err) 2430 return err; 2431 2432 err = kvm_arm_init_sve(); 2433 if (err) 2434 return err; 2435 2436 err = kvm_arm_vmid_alloc_init(); 2437 if (err) { 2438 kvm_err("Failed to initialize VMID allocator.\n"); 2439 return err; 2440 } 2441 2442 if (!in_hyp_mode) { 2443 err = init_hyp_mode(); 2444 if (err) 2445 goto out_err; 2446 } 2447 2448 err = kvm_init_vector_slots(); 2449 if (err) { 2450 kvm_err("Cannot initialise vector slots\n"); 2451 goto out_hyp; 2452 } 2453 2454 err = init_subsystems(); 2455 if (err) 2456 goto out_hyp; 2457 2458 if (is_protected_kvm_enabled()) { 2459 kvm_info("Protected nVHE mode initialized successfully\n"); 2460 } else if (in_hyp_mode) { 2461 kvm_info("VHE mode initialized successfully\n"); 2462 } else { 2463 kvm_info("Hyp mode initialized successfully\n"); 2464 } 2465 2466 /* 2467 * FIXME: Do something reasonable if kvm_init() fails after pKVM 2468 * hypervisor protection is finalized. 2469 */ 2470 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE); 2471 if (err) 2472 goto out_subs; 2473 2474 kvm_arm_initialised = true; 2475 2476 return 0; 2477 2478 out_subs: 2479 teardown_subsystems(); 2480 out_hyp: 2481 if (!in_hyp_mode) 2482 teardown_hyp_mode(); 2483 out_err: 2484 kvm_arm_vmid_alloc_free(); 2485 return err; 2486 } 2487 2488 static int __init early_kvm_mode_cfg(char *arg) 2489 { 2490 if (!arg) 2491 return -EINVAL; 2492 2493 if (strcmp(arg, "none") == 0) { 2494 kvm_mode = KVM_MODE_NONE; 2495 return 0; 2496 } 2497 2498 if (!is_hyp_mode_available()) { 2499 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n"); 2500 return 0; 2501 } 2502 2503 if (strcmp(arg, "protected") == 0) { 2504 if (!is_kernel_in_hyp_mode()) 2505 kvm_mode = KVM_MODE_PROTECTED; 2506 else 2507 pr_warn_once("Protected KVM not available with VHE\n"); 2508 2509 return 0; 2510 } 2511 2512 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) { 2513 kvm_mode = KVM_MODE_DEFAULT; 2514 return 0; 2515 } 2516 2517 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) { 2518 kvm_mode = KVM_MODE_NV; 2519 return 0; 2520 } 2521 2522 return -EINVAL; 2523 } 2524 early_param("kvm-arm.mode", early_kvm_mode_cfg); 2525 2526 enum kvm_mode kvm_get_mode(void) 2527 { 2528 return kvm_mode; 2529 } 2530 2531 module_init(kvm_arm_init); 2532